User's Guide

UNKNOWN

Reassigns low probability pixels to unknown class

Function:

Reassigns low probability pixels in each class of a classified image to the unknown class. Inputs are a classified image, the corresponding maximum likelihood/chi-square image, and a statistics file that contains the histograms of class probabilities, which are obtained as the output from the program BAYES. It produces an output classified image with those pixels with maximum likelihood/chi-square values less than a specified threshold assigned to the "unknown" class.

Parameters:

IN

Input image. This is usually obtained from BAYES and must be BYTE and single band.

INMAXCHI

Maximum likelihood/chi-square image. This is produced from BAYES. Each pixel in this image represents the maximum likelihood value/chi-square "tail-rejection" area of the corresponding input classified image pixel.

INSTAT

Statistics file that contains class histograms. This is produced from BAYES. The percentage of pixels assigned to the unknown class are computed from the histograms in this file.

OUT

Output image. Specifies the output image with pixels lower than the specified threshold assigned to the unknown class.

PIXVAL(255)

Pixel value that will be assigned to the pixels whose maximum likelihood value/chi-square "tail-rejection" area is lower than the specified threshold. These belong to the unknown class.

THRPERC(--)

Threshold percentage. This percentage will be scaled to 0-255. The image pixels with their maximum likelihood value/chi-square "tail-rejection" area lower than this value will be defined as the unknown class. If THRPERC is defaulted, then prompts will be made to get the threshold percentage for each class in the input classified image.

Examples:

1. LAS> unknown in=bayes.class inmaxchi=bayes.max instat=bayes.stat out=bayes.unk thrperc=5.0

   All the pixels in the input classified image BAYES.CLASS whose values in the maximum likelihood image BAYES.MAX are less than 13 (=0.05 x 255) are classified as unknown with a value of 255 in the output classified image BAYES.UNK.

2. LAS> unknown in=bayes.class inmaxchi=bayes.chi instat=bayes.stat out=bayes.unk pixval= 0 thrperc=2.5

   All the pixels whose values in the chi-square (tailrejection area) image are less than 6 (= 0.025 x 255) are classified as unknown with a value of 0 assigned to them. Thus, all pixels in the 2.5 percent tail-rejection area are classified as unknown.

Descriptions/Algorithm:

INMAXCHI, the second output image from the program BAYES, is either a scaled maximum likelihood image or a scaled probability image based on the chi-square values (henceforth referred to as a "chi-square image"). The parameter THRPERC is specified as a percentage. All pixels whose values in the image INMAXCHI are less than the user-specified threshold THRPERC/100 X 255 (rounded to the nearest integer) are assigned to the unknown class. Thus for the maximum likelihood image, any pixel whose maximum likelihood value is less than the specified threshold is assigned to the unknown class.

For the chi-square image, the values in the image INMAXCHI represent the scaled cumulative tail-rejection area for the corresponding pixels. For example, if all the pixels in the 5 percent tail-rejection region are to be assigned to the unknown class, the parameter THRPERC has to be set to 5. This means that all pixels whose values in the image INMAXCHI are greater than or equal to 13 (= 0.05 x 255) have a probability of at least 95 percent of correct classification.

The unknown pixels are assigned a value of PIXVAL. The number of pixels assigned to the unknown class and the percentage are printed out. The percentage is computed from the histograms in the statistics file INSTAT.

Nonfatal Error Messages:

None.

Fatal Error Messages:

1. [unknown-notalwgrp] Input image must be single band

   Use classified image from BAYES as input.

2. [unknown-unablhist] Unable to get histogram from statistics file

   The statistics file output from BAYES must be used.

3. [unknown-unablncls] Unable to get number of classes from statistics file

   The statistics file output from BAYES must be used.

4. [unknown-notablnam] Unable to get class names from statistics file

   The statistics file output from BAYES must be used.

5. [unknown-ignorwind] No window allowed--ignored

   Do not specify window for input image.

6. [unknown-ifilerr] Error opening input file

   Check input image name

7. [unknown-ofilerr] Error opening output file

   Check output file name

8. [unknown-opnstfail] Unable to open statistics file

   Use statistics file from BAYES.

9. [unknown-badinp] Error accessing input file name

   Check input file specification.

10. [unknown-badoutp] Error accessing output file name

    Check output file specification.

11. [unknown-badmin] Error accessing chi-square image file name

    Use chi-square image from BAYES.

12. [unknown-badstat] Error accessing statistics file name

    Use statistics file from BAYES.

User Note:

1. The user may use EDITSTAT or HISTPLT to display the class histograms for threshold selection.